Inspiration

Sponsorship conversations still open with the property’s logo and a guess about the room. Whether it's Chelsea FC, the LA Dodgers, a major music festival like Coachella, or a hit podcast like SmartLess: the audience already concentrates on specific brands and walks past others.

We built SponsorFit to give partnerships directors, sports agents, and brand leads a decision desk they can put on a screen in a pitch meeting — not another consumer taste quiz, and not an opaque "cultural DNA" poster. It starts from real empirical concentration and will never invent a number to make the slide look good.

The hackathon judges include people who buy and sell these high-stakes partnerships every day (such as Todd Boehly, Michael Abrams, and Jason Calacanis). The desk had to survive the real question: "Show me Chelsea, now put the Dodgers next to them, and prove why this brand fits."

What it does

You enter any sports team, creator, artist, or event, or click a 1-click scout preset. SponsorFit’s autonomous agent pipeline:

Resolves the entity: Uses Qloo's /search and /entities to load the exact cultural record.
Pulls brand affinities: Queries Qloo /v2/insights for audience affinity with explainability attribution chips.
Reads demographic & cultural profiles: Extracts signed age and gender index deviations, taste tags, and a cross-domain snapshot (associated artists, movies, and travel destinations).
Computes transparent Fit Scores: Evaluates each brand using a formulaic multi-factor score: $$\text{Fit Score} = 60\% \times \text{Affinity} + 25\% \times \text{Popularity} + 15\% \times \text{Demographic Alignment}$$
Generates an Activation Heatmap: Maps physical high-affinity geohash clusters inside the top returned destination or target activation city, filtering out remote outliers.
Audience Comparison: Overlaps shared sponsor brands and comparative taste tags between two properties (e.g. Chelsea FC vs LA Dodgers).
Drafts Anti-Hallucinated Pitches: Generates 3-angle customized sponsor briefs via LLM (AIML API), strictly protected by an evidence gate — if any number does not trace back to Qloo signals, the draft is discarded.
1-Click Export: Exports the brief directly to Clipboard, Markdown, or a print-ready single-page HTML report.

How we built it

Framework: Next.js 15 (App Router, TypeScript), Tailwind CSS, deployed on Vercel.
Data Architecture: 100% live Qloo Taste AI API calls (/search, /entities, /v2/insights, /v2/analysis/compare). No mock data or pre-baked JSON fixtures.
Streaming Pipeline: Server-Sent Events (SSE) stream 9 distinct agent cues to the front-end in real time.
Geospatial Engine: Geohash-to-coordinate decoding with universal median-distance cluster focus (focusCluster) to cleanly map activation pockets.
Safe LLM Orchestration: AIML API (gpt-4o-mini) strictly constrained by numerical grounding validation.

Challenges we overcame

Payload & Latency Budgets: Qloo's /search returns deep tag hierarchies that can exceed 2 MB. We optimized search payloads, tuned timeouts, and implemented fail-fast guards so upstream latency never blows serverless route limits.
Entity Disambiguation: Resolving cultural festivals vs municipal administrative areas (e.g. Coachella Festival vs Coachella City, California) required strict classification logic prioritizing cultural and event entities.
Geospatial Clustered Outliers: Raw heatmap signals can span wide geographical regions. We engineered an adaptive median-distance outlier filter to focus activation maps purely on dense metropolitan commercial hubs.
Eliminating AI Hallucination: Rather than letting LLMs invent sponsorship stats, fees, or metrics, every sentence in our generated pitch must trace to verified Qloo evidence chips.

Accomplishments that we're proud of

True Live Qloo Integration: Every single board, chart, heatmap cell, and score is computed live at request time.
Single-Screen Decision Desk: A complete sponsorship intelligence suite (agent trace, brand ranking, math breakdown, audience demographics, activation map, and comparison) all responsive on a single unified canvas.
Robust Quality: 18 passing automated unit tests covering scoring math, response parsing, outlier clustering, rate limiting, and hallucination gates.

What we learned

Affinity is not probability: Qloo affinity measures audience concentration relative to a global baseline.
Demographic indexes are deviations: Age and gender indexes represent signed z-score shifts rather than raw percentages.
Heatmaps are within-city activations: Heatmap geohashes answer "where inside this city do we activate?" while destination insights answer "which market should we enter?"

What's next for SponsorFit

Category Exclusions: Allow partnership managers to pin negative tag filters (e.g. gambling, alcohol, or competitor categories) via Qloo's filter.exclude.tags.
Live Deal Valuation Index: Correlate Qloo popularity and fit scores with public media value benchmarks for automated contract ballpark estimates.
Multi-Property Tour Routing: Optimize multi-city activation routes for touring artists and sports franchises.